Same Content, Different Answers: Cross-Modal Inconsistency in MLLMs

TL;DR

This paper introduces new benchmarks to evaluate cross-modal inconsistency in multimodal large language models, revealing significant modality gaps and factors affecting model reasoning across vision and language.

Contribution

The paper presents REST and REST+ benchmarks for systematic evaluation of cross-modal inconsistency in MLLMs, highlighting their limitations and influencing factors.

Findings

State-of-the-art MLLMs show substantial modality inconsistency.

Rendering text as image or vice versa does not resolve inconsistency.

Visual features like color and resolution impact model performance.

Abstract

We introduce two new benchmarks REST and REST+ (Render-Equivalence Stress Tests) to enable systematic evaluation of cross-modal inconsistency in multimodal large language models (MLLMs). MLLMs are trained to represent vision and language in the same embedding space, yet they cannot perform the same tasks in both modalities. Our benchmarks contain samples with the same semantic information in three modalities (image, text, mixed) and we show that state-of-the-art MLLMs cannot consistently reason over these different modalities. We evaluate 15 MLLMs and find that the degree of modality inconsistency varies substantially, even when accounting for problems with text recognition (OCR). Neither rendering text as image nor rendering an image as text solves the inconsistency. Even if OCR is correct, we find that visual characteristics (text colour and resolution, but not font) and the number of vision tokens have an impact on model performance. Finally, we find that our consistency score correlates with the modality gap between text and images, highlighting a mechanistic interpretation of cross-modal inconsistent MLLMs.